权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

Research and Development of Self-organizing Learning Chip and Its Application to Robotics

自组织学习芯片的研发及其在机器人领域的应用

基本信息

批准号：
14205038
负责人：
YAMAKAWA Takeshi
金额：
$ 33.53万
依托单位：
Kyusyu Institute of Technology
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (A)
财政年份：
2002
资助国家：
日本
起止时间：
2002 至 2005
项目状态：
已结题

项目摘要

(1)In order to improve recognition ability of mobile robot, a nonlinear manifold Self-Organizing Map (SOM) in which units include nonlinear manifolds instead of single vector in the ordinary SOM was proposed.(2)Batch learning algorithm of the Self-Organizing Relationship (SOR) network was proposed to improve learning convergence. Furthermore, by embedding subjective evaluation criteria of users to the SOR network, the SOR network can be applied to complex system. We applied it to orbit decision of four-wheel car and control of trailer-truck buck-up control.(3)To realize real-time processing and compactness of SOM and SOR network, SOM chip in which 25 units and 16-D reference vectors are included is implemented.(4)It is important how to get the outer world information effectively. A multi-layered basis function network model was developed to get the outer world information and its hardware implementation was done for the real-time operation. We showed that the proposed approach was usef … More ul for human-like preprocessing such as a facial feature extraction and a character area extraction in text documents. As one of robot control methods, an effective swarm behavior generation algorithm based on simple rules also has been developed using the limited sensory information.(5)The applications of the Self-Organizing Learning Chip into robotics are discussed through simulations and experiments. We focused on the unsupervised learning capability of the Self-Organizing Learning Chip, the algorithm of the chip is introduced into the decision making system of mobile robots. The results of obstacle avoidance simulations and experiments, and adaptive control show that the robots can take actions adaptively and adjust their decision making system on-line. And combining the decision making system and modular network SOM, the rapid adaptation to the change of dynamic property can be realized. The results show that the Self-Organizing Learning Chip is suitable to the realtime systems like robotic system. Less

(1)In摘要为了提高移动的机器人的识别能力，提出了一种非线性流形自组织映射（SOM）方法。（2）提出了自组织关系（SOR）网络的批学习算法，以提高学习收敛性。此外，通过在SOR网络中嵌入用户的主观评价准则，SOR网络可以应用于复杂系统。并将其应用于四轮小车的轨道决策和挂车的顶升控制。(3)To实现了SOM和SOR网络的实时处理和紧凑性，实现了包含25个单元和16维参考向量的SOM芯片。(4)It如何有效地获取外部世界的信息是很重要的。提出了一种多层基函数网络模型来获取外界信息，并对其进行了硬件实现以保证实时性。我们表明，所提出的方法是有用的， ...更多信息用于类似于人类的预处理，例如文本文档中的面部特征提取和字符区域提取。作为机器人控制方法之一，基于简单规则的群体行为生成算法也已被开发出来，利用有限的传感器信息。（5）通过仿真和实验研究了自组织学习芯片在机器人中的应用。重点研究了自组织学习芯片的无监督学习能力，并将其算法引入到移动的机器人的决策系统中。避障仿真和实验以及自适应控制的结果表明，机器人能够自适应地采取行动，并在线调整其决策系统。并将决策系统与模块化的SOM网络相结合，实现了对动态特性变化的快速适应。结果表明，自组织学习芯片适用于机器人等实时系统。少